Key Technologies Of Failure Analysis And Reliability Prediction Based On Statistical Learning Theory

Reliability prediction based on effective identification and analysis of failure is very meaningful to the enterprises in failure prevention and maintains scheduling. With the improvement of product reliability, it is difficult to acquire enough data used to identify failure based on data analysis. Technologies of failure analysis and reliability prediction based on statistical learning theory are presented. A prototype system is developed for validation.The framework structure of failure analysis and reliability prediction system is built based on the analysis of work principle.The statistical learning theory and support vector machines are introduced to identify the mechanical product failure, which is a typical small-sample problem. The binary classification method and the multi-classification method are applied in the classification of single failure and multi-failure mode in automobile engine. The result shows that the statistical learning theory and support vector machines can improve the accuracy compared to the conventional machine learning methods.The FTA and FMEA are integrated to analyze the failure modes identified by the support vector machines. The new method can lessen the limitation of FMEA or FTA. Meanwhile, the Fault Tree (FT) can be generated automatically by the FMEA tabulation, which is help to effectively identify the specific cause of failure.The linear regression method of degradation prediction based on the statistical learning theory is presented. The degradation data is used to predict the reliability of product with related degradation modes. A case study is given taking the shell of the automobile engine as an example. The result shows that this method can accurately judge whether the reliability of the product is lower than the customer's expectation.A failure analysis and reliability prediction system is developed to validate the theory and methods mentioned above.